Bet Builders and Correlation: Why the Legs Do Not Multiply

A bet builder is not a small accumulator

An accumulator combines selections from different matches. A bet builder — a same-game multi, a request-a-bet — combines selections from inside one match: a team to win, over a number of goals, a named player to score, a count of corners or cards. The betslip looks the same and the settlement rule is the same, every leg must win, so it is natural to assume the pricing is the same too.

It is not, and the reason is not commercial. Multiplying two prices together is arithmetic that is only valid when the two events are independent — when knowing that one happened tells you nothing about whether the other did. Across two different matches on a Saturday afternoon that assumption is roughly true, which is why our accumulator guide can treat the compounding of margin as the whole story. Inside a single match it is false, and often badly false. The goals that decide whether a match goes over 2.5 are the same goals that decide whether both teams scored.

So the question is not whether a builder price differs from the product of the singles. It always will. The question is by how much it should differ, and whether the difference in front of you is bigger than the football justifies.

One number: the lift

Take two legs, A and B. Write the chance of each as a probability, and write the chance of both landing together as a third. If the legs were independent, the third would be exactly the first two multiplied. Divide what it actually is by what independence would give, and you have a single number that measures the whole relationship:

The fair builder price follows directly: it is the product of the single prices divided by the lift. That single line is the whole of what a correlation model does, expressed in a form you can use without one.

The table: what the lift actually is

The figures below come from the standard model of football scoring: each side scores goals at its own average rate, independently of the other, and the match result and the goal totals both fall out of the same two rates. The method is set out in full at the foot of this page. The only input is how many goals the match is expected to produce in total, and you can read that off the over and under market rather than guess it — the second column gives the chance of over 2.5 goals that corresponds to each total, and our odds converter turns a price into that percentage.

The result worth stopping on is not any single figure but what is missing from the table: the teams. These lifts were calculated for an evenly balanced match, and then recalculated with the same expected total shared as unevenly as 78 to 22 between the two sides — a thorough mismatch. Across that whole range the both-score-and-over figure moves by about 0.02, and none of the four columns moves by more than 0.04. Correlation between goal-based legs is a property of how many goals the match is expected to produce, not of who is expected to win it. You do not need a view on the teams to use this table; you need a view on the total, which the market will give you.

At 2.8 expected goals, a figure close to what a typical Premier League fixture is priced around, the four pairs translate into fair builder prices like this:

The one that runs backwards

Two of the four pairs above have a lift below 1.00, and one of them is built constantly. Backing a side to win and both teams to score feels like a natural pair: you are asking for an entertaining match involving the team you fancy. The football says otherwise. A clean sheet is one of the most common ways a side wins a match, so every scenario in which your team keeps one satisfies the first leg and destroys the second. At 2.8 expected goals the lift is 0.89, which means the fair price is about 113% of the product of the singles: the multiplication underpays.

The same is true, more strongly, of a win combined with under 2.5 goals: a lift of 0.83 and a fair price around 121% of the product. Notice also that the negative pairs behave in the opposite direction to the positive one as the match gets higher-scoring. The more goals a match is expected to produce, the weaker the both-score-and-over relationship becomes, because both legs are drifting towards being likely anyway; and the stronger the win-and-under conflict becomes, because a low-scoring outcome grows rarer.

None of this is an argument that a negatively correlated builder is a good bet. It is the reason a builder cannot be judged by whether its price looks generous next to the singles. On half the common combinations, generous is exactly what a fair price should look like.

The test: what the operator has actually applied

You cannot see an operator's correlation model. You can see the number it produced. Take the same selections as singles, multiply their prices, and divide by the builder price:

A worked example, with round numbers chosen to keep the arithmetic visible rather than to describe any real market. Both teams to score is offered at 1.80 and over 2.5 goals at 1.90. The product is 3.42. The builder on the two together is offered at 2.20. The applied factor is 3.42 ÷ 2.20 = 1.55. The over 2.5 price implies a little over 52%, which puts the match near 2.8 expected goals, where the table gives a lift of 1.49. The operator has priced 1.55 against a structural 1.49: the correlation is doing almost all of the work, and the gap of roughly 4% is what has been added on top of whatever margin the two singles already carried.

Two cautions on reading that gap. First, it is a comparison against a model, not against the truth, and a builder that includes a player or card leg sits outside what a goals model can describe. Second, it is not the total cost of the bet. It is the extra, on top of the overround already inside each single price. Our margin calculator measures that underlying overround on any market, and the accumulator calculator and bet calculator do the multiplying so you can compare the two figures directly.

What this does not fix

The margin still compounds. Every single price contains an overround, so multiplying two of them carries that overround twice over: at 5% a leg, the product is already 10.25% over before a correlation adjustment is applied at all. The full compounding table by number of legs is in the accumulator guide. A builder does not escape it — it hides it, because the operator publishes one price rather than several, and correlation and margin arrive inside the same number.

Settlement still governs the outcome. A builder leg that turns on a count of goals or corners is settled on ninety minutes plus stoppage time on standard football markets, which is why a knockout tie that reaches extra time can pay a builder and lose the bet on who went through — set out in Extra Time and Penalties. And a builder built after kick-off is subject to the bet delay and market suspensions covered in our in-play betting guide.

For the competition most of these bets are struck on, our Premier League 2026/27 betting guide tracks the season's confirmed dates and how its season-long markets settle.

Frequently asked questions

Method and sources

Every figure in the tables was computed rather than collected. Each side is modelled as scoring goals at its own constant rate, independently of the other, which is the standard starting point for football scoring and gives a Poisson distribution for each side's goals. For a given expected total, the full joint distribution of scorelines was enumerated to 17 goals a side, and from it the probability of each leg, the probability of both legs together, and their ratio. The stability figures come from repeating the whole calculation with the same expected total split between the sides at 50/50, 55/45, 60/40, 65/35, 70/30, 75/25 and 78/22. Any reader with a spreadsheet can reproduce the table.

The model's known limitation is stated rather than hidden. Independent Poisson slightly understates draws and low-scoring results in real football, the correction usually attributed to Dixon and Coles; it touches the low-scoring cells, so the win-and-under and draw-and-under relationships in particular should be read as approximations, while the both-score-and-over figures are barely affected. Settlement descriptions reflect standard UK and Irish bookmaker practice, and an individual operator's published rules always take precedence over a general description. Prices used in the worked example are illustrative round numbers, not quotations from any market.